Interrussia Fellowship June 2026

Quantum-Enhanced
Neural Detection.

Closing the <3mm sensitivity gap. Our hybrid quantum-classical network captures non-local spatial correlations to detect micro-metastases early.

Test the Architecture Read the Pipeline
Comparative Benchmark

Experimental Pipeline

Our rigorous testing pipeline evaluates classification accuracy, structural parsing capacity, and noise robustness limits to benchmark quantum capabilities against state-of-the-art classical convolutional baselines.

Experiment 1 Detection

Tumor Detection

Validate basic diagnostic capability of the model on raw, binary input scans.

Input MRI Slice
Output
0 ➔ No Tumor 1 ➔ Tumor
Target Metrics
Accuracy Precision Recall F1 Score AUC Sensitivity Specificity
Purpose

Validate basic diagnostic capability.

Experiment 2 Structural

Tumor Size Classification

Evaluate whether the model captures clinically meaningful tumor characteristics and boundary volumes.

Input MRI containing tumor
Output
0 ➔ Small Tumor 1 ➔ Large Tumor
Target Metrics
Accuracy Precision Recall F1 Confusion Matrix
Purpose

Evaluate whether the model captures clinically meaningful tumor characteristics.

Experiment 3 Stress Test

Noise Robustness

Stress-test architectures under varying simulated scanner noise levels and motion artifacts.

Noisy Datasets Evaluated
Original 5% 10% 15% 20% 25% 30%
Noise & Blur Types
Gaussian Salt-and-Pepper Rician (MRI-spec) Motion Blur Gaussian Blur
Evaluation Flow
Classical CNN ➔ Hybrid Model ➔ Compare Degradation
Purpose

Compare performance degradation and robustness thresholds.

5
Benchmarking Matrix

Model Comparison

Direct performance breakdown comparing classical, pure quantum, and quantum-classical hybrid networks.

Architecture Accuracy & F1 ROC AUC Sens & Spec Resource Footprint Robustness to Noise
Classical CNN 88.3% / 87.5% 0.942 87.0% / 89.6% Params: 23.5M
Inference: ~600ms		 Moderate degradation at >15% noise levels.
Quantum CNN 85.1% / 84.8% 0.910 83.2% / 86.8% Params: 12.4K
Inference: ~1200ms		 High robustness; features invariant to local perturbation.
Hybrid CNN + VQC Best 94.8% / 94.2% 0.978 95.2% / 94.4% Params: 15.6M
Inference: ~800ms		 Strong robustness; maintains >90% accuracy up to 30% noise.
ion>

Inference Dashboard

Upload a 128x128 grayscale MRI scan or choose an evaluation template to run classification.

Patient Details (Optional)

Analysis Output No Scan Staged

Ready for pipeline execution

Stage up to 3 MRI scans and select neural network architecture to execute the comparative benchmarking pipeline.